Alphabet, Google’s parent company, has revealed its latest endeavour to make science fiction into reality—The Everyday Robot Project.

Led by X Company, the Everyday Robot Project has been underway for years with the “moonshot goal” of creating a general-purpose learning robot. As modern-day robots are expensive and geared towards industrial means, X Company is hoping to create an affordable robot that can be taught to help anyone.

According to the company, constructing robots to function in unpredictable, constantly changing human environments is a complex feat. While people can combine several skills, such as understanding, acting and navigating to accomplish a goal, robots generally need a lot of coding to do the same.

“For robots to be useful in everyday environments we need to move away from painstakingly coding them to do specific and structured tasks in exactly the right way at exactly the right time,” wrote Alphabet X robotics project lead Hans Peter Brondmo in a Medium post. “We have concluded that you have to teach machines to perform helpful tasks; you cannot program them.”

In the early days of the initiative, back in 2015 and 2016, X Company collaborated with Google AI, ultimately showing that robots could learn how to carry out a task. By human demonstration, from shared experience and simulating robots in the cloud, the team investigated how to train robots to function in the real world.

Over the past few months, X Company has been running an experiment for the robots to perform a useful task: sorting trash in workplace environments. The robots set out to divide cups, bottles, snack wrappers and more items across landfill, recycling and compost bins.

“Everyone has put waste into the wrong bin at some point, whether as an ‘oops’ or because our mind was somewhere else,” Brondmo explained. “As a result, contaminated items that could have been composted or recycled have to be sent to landfill. In a typical office, no one sifts through items to check for contamination, so this felt like a valuable problem to put our robots to work on.”

Several machine learning techniques were used to teach the robots, such as simulation, reinforcement learning, and collaborative learning. Every night, thousands of virtual robots practice sorting in a virtual environment in a cloud simulator. Later, real robots were trained to fine-tune their sorting skills. The real-world training and the simulated training data are then integrated and shared with the rest of the robots, teaching them all in the process.

The X Company robots are now capable of sorting trash with a waste contamination level of under five percent—a fraction of the 20 percent shown by humans. Brondmo explained that these results are promising for two main reasons. One, the team was able to create a system that uses a robot’s abilities to do something useful. Two, the robots demonstrate that it’s possible to learn real-world tasks by just practicing—no extensive coding required.

“It will be years before the helpful robots we imagine are here, but we’re looking forward to sharing more robot adventures along the way.”